Core Concepts

Before building with agents, it helps to have clear mental models for the key pieces. Think of these concepts like understanding the parts of a car before you learn to drive.

Understanding these concepts lets you:

• Choose the right tool for a given problem instead of guessing
• Debug faster — when something breaks, you'll know which layer to look at
• Compose systems confidently — knowing how pieces fit together makes you a better architect
• Communicate clearly with teammates, docs, and AI assistants

The five core concepts

Concept map — how they relate

Every agent system is built from these five pieces, each playing a distinct role:

A minimal agent needs only a model and a goal. As you add memory and tools, the agent becomes more capable. The harness is what ties it all together into something you can actually ship.

The three-layer mental model

Every AI agent system has three layers:

┌─────────────────────────────────┐
│         YOUR APPLICATION        │  ← What users see and interact with
├─────────────────────────────────┤
│         AGENT HARNESS           │  ← Orchestration, memory, tools, routing
├─────────────────────────────────┤
│         AI MODEL                │  ← The reasoning engine (Claude, GPT, etc.)
└─────────────────────────────────┘

You don't have to build all three yourself — that's what the tools in this cookbook are for. But understanding what lives at each layer makes you dramatically better at both using and building agent systems.

Where to start

If you're new to agents, read the concepts in this order:

1. Models — understand what the model actually does
2. Agents — see how a model becomes an agent
3. Tools — learn how agents take action
4. Memory — learn how agents maintain context
5. Harnesses — see how it all gets wired together

Once you have a feel for these building blocks, head to the Recipes section to see them in action.